Next invitations
Proposes trays for next week with a reason grounded in the logs.
Coach
AI with a leash
The coach reads your logs, not the internet, and tells you which entries each suggestion is based on. Every feature has its own switch, and names are replaced with placeholders before anything leaves our servers.
What it does
Coaching point
Drafts one kind point for the adult on duty, in their language.
Reading prompts
Eight CROWD prompts for any book title or page photo, in the language you ask for.
Translation
Quotes and scripts between the family’s languages.
Evidence questions
“What does the evidence say about X?” answered from the curated research with the same steelman-and-critique balance.
Balance drift
“No Malay logged in nine days.” Calm, once.
Guardrails
What it will not do
Never diagnoses
It does not label the child. Red-flag checks are rules, not the model.
Never recommends rewards or punishments
The playbook forbids them; so does the coach.
Always cites
Every suggestion lists the entries it used.
A switch per feature
Translate, book prompts, weekly suggestions, evidence answers: each is a switch in Settings › Privacy. Turning the coach off deletes its stored chats and suggestions.
Audited
Every call is written to an audit log the family can read.
Minimal data
Only the text needed for the feature is sent.
FAQ
Which model does it use?
Claude models from Anthropic, with a faster model for translation and prompts and a stronger one for suggestions. The provider can change; the guardrails and the audit log do not.
Does it see photos?
Only for the reading-prompt feature when you give it a page photo, and only if that feature is on.
Run the day. Keep the year.
Create the family and invite the other adult. Free during early access.